On-chain systems fail on the boring parts. A reorg handled optimistically. An RPC node that answers confidently with stale state under load. Decimals assumed rather than read. A payout path with no idempotency key, so a retry pays twice.
An indexer that drifts a few blocks from chain state and never says so. None of it shows up in a demo. All of it shows up in reconciliation, weeks later, as a number that does not match.
DeFi development services here mean protocol integration, on-chain data and indexing, settlement and reconciliation paths, and a simulation layer so a change can be evaluated against forked mainnet state before it touches a live chain.
Rust where correctness and latency justify it, not as a badge on a services page.
Anupam Shah ran payment infrastructure at $60M+ volume in MAS-regulated, SOC 2 / ISO 27001 environments, and that work teaches the same lesson a reorg teaches, only with a chargeback attached.
On-chain work is the same discipline with worse recovery: off-chain, a bad write can usually be reversed quietly.
On-chain, the mistake is public, permanent, and somebody else's argument by morning.
Security audits are referred to independent auditors. Token design, fundraising, market making, custody of client funds and financial advice sit outside the engagement; the strategy and the outcome stay yours, and we publish no performance figures.
When a DeFi development company puts a yield number on its own services page, the useful question is whose capital produced it and who was carrying the risk while it did.
Scope
We walk every path that moves value and write down the assumptions holding it up: finality treated as instant, a reorg depth nobody chose, an RPC provider trusted to stay current under load, decimals hardcoded, a retry with no key behind it.
The list that comes out is the scope.
Smart contract integration against contracts you do not control, and on-chain data pipelines built to converge back on chain state rather than drift away from it. Upgradeable proxies change their ABI without asking.
Backfills have to be repeatable on a bad day, not just on a clean one. An index that cannot be rebuilt from block zero is a cache pretending to be a database.
Every path that moves value is safe to retry, with reconciliation and alerting designed in rather than bolted on after the first mismatch. A retried payout is a double payout.
On-chain the same bug arrives wearing a different hat — a reorg, a stuck nonce, a transaction that landed while your service was restarting.
Chain scope is where this work quietly doubles. "And also on Base" is a second integration with its own reorg behaviour and its own reconciliation, not a checkbox.
A forked-state simulation harness, a runbook for the failure modes the review turned up, and alert thresholds that fire before reconciliation does. You should be able to evaluate your next change yourself.
Nothing in the delivery depends on a service we host, a key we hold, or a licence we control.
A fixed fee against written scope, agreed before work starts and billed at milestones as the work lands.
You do, outright. IP assignment is signed before repository access, by everyone who touches the code, and it covers AI-assisted output. Handover includes the runbook and the rollback path. Nothing depends on a licence we control.
Nobody before an NDA. Access is read-only until scope is signed, and you can revoke it at any time. The AI tools used on your code, and what each one retains, are listed on the governance page.
Founder, Provingmark
Thirty minutes, no charge.
Related: Hire blockchain developers · Execution, risk & settlement systems · AI agents for on-chain systems · AI consulting & implementation · How we scope and bill